The ggplot2 Grammar

ggplot2 is the most influential R package of the last 20 years. It is based on a 1999 book, The Grammar of Graphics, by Leland Wilkinson. The grammar's central claim: every chart, no matter how complex, is built from the same handful of components.

Learn those components and you can build any chart you can imagine.

The components:

Each part is added with +.

A first ggplot

Read it as a sentence:

• Start a plot of mtcars (the data)
• ...mapping wt to x and mpg to y (the aesthetics)
• ...and add a layer of points (the geometry)

That's the whole grammar. Every ggplot in existence is some extension of this skeleton.

Geometries: change the chart, not the rest

Try different geoms on the same data:

You change one word (geom_point → geom_smooth) and you get a fundamentally different chart. Same data, same mapping, different mark.

A small map of the most-used geoms:

Mapping vs. setting

This is the single most common confusion for ggplot beginners.

• Inside aes(): you're saying "let this column control this property."
• Outside aes(): you're saying "set this property to this constant value."

If you put a literal color = "steelblue" inside aes(), ggplot will treat "steelblue" as a category name and give you the default categorical color, with a useless legend. The takeaway:

• Want it driven by data? → put it inside aes()
• Want it constant? → put it outside aes()

Faceting: small multiples for free

Often you want the same plot, broken into panels by some variable. ggplot makes this trivial with facet_wrap() and facet_grid():

One line — facet_wrap(~ Species) — and you get three side-by-side panels, one per species. Small multiples, the single most powerful visualization technique for spotting group-level differences.

Labels and titles

labs() is for everything textual:

Always set the title and axis labels. Always. They cost almost nothing and save your reader a lot of guessing.

A few more geoms in action

A bar chart with counts (categorical x):

A grouped boxplot (numeric y by categorical x):

A line chart over time:

Notice how almost identical the structure is across very different chart types. That's the grammar at work.

Themes

A theme() call lets you tweak almost any visual element. Ggplot also ships with several full themes:

For most analytical work, theme_minimal() is a great default — clean, gridded, modern.

Test your understanding

QuestionSelect one

What's the difference between writing geom_point(color = "red") and geom_point(aes(color = "red"))?

Hint: think about what aes() is for — mapping data to a property versus setting a property to a fixed value.

They are equivalent.

The first errors; only the second works.

The first sets every point to red. The second maps the literal string "red" to color as if it were a category, and ggplot picks the default color for that category — usually not red — and adds a useless legend.

The second is faster.

QuestionSelect one

Which is the right ggplot to make a scatterplot of hp (x) vs mpg (y) from mtcars?

ggplot(mtcars) + geom_point(hp, mpg)

ggplot(mtcars, aes(x = hp, y = mpg)) + geom_point()

plot(mtcars$hp, mtcars$mpg)

ggplot(mtcars, hp, mpg) + scatter()

QuestionSelect one

To split a ggplot into one small panel per Species in iris, you add:

facet_wrap(~ Species)

group_by(Species)

color = Species

split(Species)

Mini challenge: a publication-ready iris plot

Build a ggplot of iris showing Sepal.Length on the x axis, Petal.Length on the y axis, colored by Species, with a linear smoother per species, faceted by species, with a real title and axis labels.

We can now make principled, well-labeled charts. The next page is about reading them — what to look for when a chart lands on your desk.